Hybrid Principal Component Analysis Using Boosting Classification Techniques: Categorical Boosting

Our study focuses on the critical function of wearable sensors, particularly inertial-based sensors, in managing the everyday physical activity of humanoid humans across a wide range of applications, both present and future. Existing classifiers are limited in their ability to handle the intricacy of difficult categorization tasks. In response to this difficulty, our research underlines the value of physical activity recognition, which includes actions such as roping, walking, and jumping. To address this, we propose a classification model that employs CatBoost, a form of boosting techniques, as well as hybrid principal component analysis. When compared to existing models, our suggested PCA+CatBoost classification model outperforms them all. We ran benchmark tests on the widely available PAMAP2 dataset, challenging the model to recognise and assess the intensity of a variety of physical activities. The results show that our procedure is effective, with an incredible accuracy of 99.93%.